This checkout had no git history at all, so this one commit carries the whole
of the Everwood work as a single import on top of chairmanpow/benilla-octo's
`main` (706cd71). The two trees have genuinely diverged — each has files the
other does not — so this is a fork point, not a catch-up.
Today's work, which is the reason a baseline exists at all:
- Booth previews no longer cast world shadows (`NO_BOOTH_SHADOW`). Portrait
characters were entering the sun's shadow cascades, so every skinned-rig
shadow specialization compiled live mid-frame; the hitch detector logged
"main thread blocked 1316 ms" beside a `compiled LIVE` warning. This was
the Fashion/Tmog stutter.
- Per-pane dress-up: each `DressUpModel` pane assembles and stages its own
tried-on look instead of every pane mirroring the live player.
- Item template cache is append-only JSONL written on the IO pool, replacing
an 11.6 MB whole-map JSON rewrite on the main thread every 10 s.
- `WOW_SYSTEM_TIMING` rebuilt per-thread: it took 2 mutexes per system enter
and 4 per exit, costing 15 ms a frame (590 fps -> 60) and corrupting the
measurements it was taken to produce.
- The visibility-churn sweep moved to its own `WOW_FLASH_CHURN` switch, so a
frame-cost reading no longer pays 2.7 ms a frame for it.
Local state (`benilla-config/`, `logs/`, `target/`) is excluded by the
existing .gitignore; the account credential it holds is not in this commit.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01SgstFWtn1RxEJ5kVrpgYwn
The fourteen 1.12.1 functions #336 lists are defined, each derived from its reference address, and benilla's layer drops the two places it stood in for the stock bag code.
- **Keyring:** `KeyRingButtonIDToInvSlotID` (`0x4c8150`) exists, and the keyring slots answer as inventory slots 82-113. The stock `ContainerFrame.lua:616-619` hover path now runs unchanged, and the layer's keyring `OnEnter` wrapper is deleted.
- **Bags:** the layer no longer replaces `CloseAllBags`. The stock one (`ContainerFrame.lua:702-707`) closes the backpack only when it was not open before the vendor or mailbox opened it (#541). The deviation that opens every bag stays, and its comment now names its only callers.
- **Inventory:** `ShowInventorySellCursor`, `SetInventoryPortaitTexture` and `UpdateInventoryAlertStatus` exist.
- **Character and pet:** `GetDamageBonusStat` (ChrClasses field 2), `GetPetTimeRemaining` (armed from `SMSG_PET_SPELLS`) and `GetActionAutocast` exist.
- **Mail:** `ClearSendMail`, `GetNumPackages`, `GetPackageInfo` and `SelectPackage` exist, over a new `Package.dbc` loader. A selected package rides in the mail send. The compose reset now also clears the stationery, the package and the attachment, as `0x4acdc0` does.
- **Tabard:** `GetTabardInfo` opens the tabard designer through the registrar (`0x5e00e0`).
- **World map:** `GetWorldLocMapPosition` projects a map position onto the displayed map (`0x4a7360`).
- **Loot:** `SetLootPortrait` binds a unit loot source's portrait (`0x4c2b40`).
Verified:
- A test per function for its shapes, raises and effects, failing on main.
- A keyring hover test through the stock path, failing on main.
- A backpack-over-vendor test, failing on main.
- `scripts/gates.sh` and `scripts/smoke.sh` are green.
- A live A/B run: hovering a key asked for `player:82` and showed the key's tooltip. After a vendor, opening the Help panel left the backpack open and closed bag 1; on main the backpack closed.
Built on macOS; no platform seam touched.
Fixes#336Fixes#541🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The strings pass for `benilla-app`, the last crate: string literals and test names lose the private record numbers and the workspace's words that a comment pass cannot touch.
Fixes#92. Fixes#36.
- **Messages.** About 230 assertion, panic, `expect` and log messages, printed tool labels, and comments inside embedded Lua fixtures drop their record numbers; the eight that named "the director" keep their fact in plain words. Stale claims in a few messages are corrected on the way (a cast-bar step, a font line, the every-shot recompute).
- **Names.** Seven test and helper names that carried a record number or "director" now say what they check. `git grep` found no other use of any of them.
- **Player files.** The key-bindings file's header and `config.toml`'s header each lose a record citation. Neither header is compared when the file is read back: the bindings reader skips every `#` line, and `config.toml`'s header is a TOML comment.
- **Fixtures.** Test fixtures use neutral character, account and frame names.
- **#36.** The quest-material test labels GameObject type 10 GOOBER (in 1.12, 3 is CHEST, 9 TEXT, 10 GOOBER); the code under test was already right, and the comments were fixed with the comment pass.
- **Verification.** With comments stripped and every literal's contents masked, each file is identical to main and keeps its literals' line counts, apart from the renamed functions and one rustfmt reflow. `cargo fmt --check` is clean.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Part of #41: the second of five `benilla-app` pieces. This one is the UI windows: chat, the NPC windows (gossip, merchant, trainer, taxi, bank, mail, auction, quest giver), the unit, party, raid and aura feeds, the action bar, bags, tooltips, macros, the spellbook, the world map and the smaller panels. Their comments are rewritten to the rule in `docs/METHOD.md`, which is what the code does and the reference fact behind it, in a line. 26,648 comment lines become 9,221.
Fixes#98.
- **The mechanical strip.** The citations that were only a private record number go first, and each edit is shown to be a pure removal.
- **Wrong facts corrected.** The rewrite corrects the comments that stated wrong facts. Among them:
- Several client addresses carried a `0x400000` slip (`UPDATE_MACROS` is `0x852460`, not `0x452460`; the macro store's globals likewise).
- `QUEST_WATCH_UPDATE` carries the quest-log row (`0x4df880` scans the log table at `0xbb71c0`), not a watch-list position.
- The stock `ChatFrame_DisplayTimePlayed` now ships, so benilla's own `/played` lines duplicate it (#194); trainer refusals are logged, not shown on the error line; the player's cast result is `SMSG_CAST_RESULT`.
- Stock FrameXML and `Blizzard_*` line citations are checked against the stock 1.12 files, and vmangos line citations against the vmangos revision nearest the one the local server runs.
- Dozens more turned up on the way: docs sitting on the wrong item, removed functions, stale counts, broken intra-doc links.
- One test comment in `ui_script/questlog_tests.rs`, which landed with #184, carried the same watch-index slip and gets the same fix.
- **Deviation labels.** "Deviation:" now marks only a choice with a reason. An unbuilt behaviour, or one whose reference side is untraced, says so plainly.
- **Verification.** Every file strips to identical code, and `cargo fmt --check` is clean. Six second readers compared every file old against new, restored the facts the rewrite had dropped, and checked each new citation against the code, the stock files, vmangos and the reference.
Behaviour differences the rewrite surfaced are filed as their own issues (#188, #204, #222 and #259 among them). The comments here state what the code does. String literals are untouched; they are #92.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Comments that pointed into the maintainer's private reverse-engineering notes (a note by name, a
path, a section or finding id, the analysis's own name for a client function, a process word) now
carry the fact itself, cited by the reference client's address, by vmangos file and line, or as
"the reference". Nobody outside could follow those pointers.
Comment text only, plus the test-assert and log strings that carried the same pointers. No claim
changes: a comment the notes have since corrected keeps its wording here, and correcting it is
separate work.
The two reference generators, `scripts/gen-reference-globals.py` and
`scripts/gen-reference-verb-events.py`, named the private notes' layout to find their inputs. They
now take the input files as arguments (`--capture`, `--framexml`, `--names`, `--extents`,
`--disasm`). Run on the same inputs, both write the committed tables row for row; only the two
header lines that named the old environment variable change.
Checked per file:
- the code is identical: each file with its comments stripped compares equal before and after; a
changed string literal is compared under a mask that proves only literal text moved and every
literal kept its line count;
- no client address the old text cited was dropped;
- a second reviewer checked every added address against its derivation;
- a detector for the notes' names, paths, ids, function names and process words finds none left;
- `scripts/gates.sh` is green.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
465 tests loaded the stock interface off the install without declaring
it, and 25 addon tests assumed the install beside the corpus; each now
opens with `wow_data_or_skip!`. The gates run the workspace tests a
second time with `WOW_DATA=`, so a test that reads the install
undeclared fails at land instead of on a contributor's clone. The skip
log writes each line in one call so parallel tests do not interleave.
Verified: the workspace tests with `WOW_DATA=` in a checkout that has
the addon corpus pass 6972, fail 0, and log 1555 skips; on a fresh
clone of main before this change they failed 465.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
- the CVars are one engine-side table that outlives every interface reload: a change fires its observer, the eight latched rows wait for RestartGx, and /console gains set, cvar_reset, cvar_default, cvarlist and help (2303)
- every push into the interface runs in one feed phase after the net drain and before the tick, held by a structural test on the built schedule (2304)
- subsystems own their packet handlers, dispatched in packet order; the auction house is the first family out of the 273-arm drain (2305)
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
- a panic writes benilla-config/Diagnostics/crash-<time>.txt with the build id, the location, a backtrace and the last 200 log lines, and names the file as its last line, so a bug report can attach it (2266)
- every wire-count allocation hint is capped and the compressed update object reads through the cursor, so a malformed packet can no longer ask for gigabytes (2266)
- the 23 in-world run conditions are one set, the 45 probe variables one registry, the install-gated tests say when they skip, and every gate is timed (2266)
- one workspace-level lint allowance replaces 413 per-site attributes, and the decision corpus gets a lint in the gate (2266)
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
- the chat window is the reference's own `ChatFrame.xml` and `FloatingChatFrame.xml`, run off the player's chain; our 2216-line transcription is deleted, the feed fires `CHAT_MSG_*` and the reference's `ChatFrame_OnEvent` prints — no app-side composition into the windows, no routing table, no whisper chime or flash of ours, and the transcribed edit-box machine (sticky type, tell ring, Tab cycle) retires with it; the verbs the two files call are built to the bytes — the whole `0x98` chat-window record and its eight verbs, the 94-entry chat-type colour registry, the language walk, the channel verbs as a command queue, the slash handlers' emote/roll/assist/uninvite/console/logging verbs; `chat-cache.txt` is written and read in the reference's whole grammar, the 68-entry message-group table in flag order and the no-file boot included, and the chat bindings and macro lines run through the reference's own arms (1948)
- the spellbook is the reference's own `SpellBookFrame.xml`; our 965-line transcription is deleted, and two quirks it reproduced by hand (the pet page ignores its page number, its arrows write the spell book's counter) now come from the reference's lines; one adapter remains because the reference's autocast shine is a `<Model>` that draws no pixels in this engine — the script puts the token-texture shine beside each button's model and carries it with the model's shown state; the nine spell-slot bindings (`GetSpellName`, `CastSpell`, `PickupSpell` and their kin) now parse their arguments the way the reference's shared marshaller `0x4b3ec0` does — a numeric arg 1, a mandatory arg 2, the index gated to [0, 0x400) with "Invalid spell slot in <Verb>" on a miss, "pet" matched without case — where ours raised mlua's own message on a negative id, answered nil for 0 and above 1024, and read the wrong book for "PET" (1952)
- a long-hidden world model part parks its `Mesh3d`: the entity path keeps the batches the retained pass declines resident — 5.3k hidden submeshes at any instant at the Stormwind auction house — and Bevy's sweeps walk every `Mesh3d` whether it draws or not, so a part parks after 30 consecutive hidden frames and unparks the frame its verdict turns visible; every world model part takes part, units' bodies included, while a booth's mirror parts stay out; interleaved pairs read an alone Stormwind 0.1–0.4 ms lighter, and `WOW_NO_MESH_PARK=1` is the A/B; the audio thread's realtime class was the standing suspect for the raid's `cpu_p99` spikes with sound on, and `WOW_AUDIO_NO_RT=1` across two interleaved Blackrock pairs read no difference, so the lever stays as an instrument (1949); and a no-merge prop batch now says which rule declined it — animated (1551 at the auction house) or transparent, not order-free (340) (1951)
- the probe reads CPU per thread across its window (user and system split, because a user-mode sampler never sees kernel time) and the main thread's CPU inside the app's frame beside its whole-frame CPU; together they place the vsync tax every player pays by default — cpu_ms 10.7 uncapped against 12.5 under vsync, alone in Stormwind, with no system growing and no thread spinning: it lands on the main thread, in user time, inside the schedules, the same systems each running slower, and a low-priority spinner holding one core busy erases it to the last decimal (vsync + spinner 10.18 against uncapped 10.17), which is frequency scaling — under vsync the cores idle deeper between frames and the next burst starts at a lower clock; nothing in the client is the cause and nothing in the client changes for it (1950)
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
- spells shake the camera: `SpellEffectCameraShakes.dbc` loads beside `CameraShakes.dbc` as the group table, `SpellVisualKit` field 14 is the shake column (58 of 1772 kits carry one — War Stomp, Thunder Clap, Kodo Stomp, the ~50 Knockdown/Slam/Charge impacts, every summon, Sapper Charge, Earthquake, Sapphiron's death), and the kit shake fires exactly once per play from the first effect node's one-time arm pass (`0x620e11`), the same leg that plays the kit sound one field over — the older "fires only if nothing else did" reading would have silenced every kit with a populated effect slot; `$SHK` markers fire from GameObject and DynamicObject hosts only (the titan chest, the bomb wagon); one deliberate deviation — the reference reads a bone-attached node's position twelve instructions before it is written, so 26 of its 58 shake kits shake at the world origin and are culled by the 80 yd cut, which is an uninitialised field, not a design; 21 of 24 presets now have a live producer, up from 6
- the sound keeps playing when the window loses focus, if you want it to: 1.12 mutes everything on `WM_ACTIVATE` and offers no way out, so benilla ships that as the default (`Sound_EnableSoundWhenGameIsInBG` "0") and puts an Audio-page checkbox on it — a gate appended LAST in the main-track chain, downstream of every tap, so an unattended capture still records; wow-re's round settled that music is not exempt and that it is a mute, not a pause
- a building's far side no longer draws through its own walls: the camera-inside leg has three passes and we had one and a half — Pass 2 admits an exterior-shell group only against a frustum clipped to ONE deferred window (we marked every such group visible against the full view, under a comment calling it a benign over-draw), and Pass 3, which we had none of, draws the 24 `0x10000` groups (14 in Stormwind, 6 across Dire Maul, Razorfen's four 450–520 yd shells) from anywhere, frustum-tested on both legs by byte-identical setup; there is no antiportal branch in 5875 and no such flag in its data (815/815 roots and 5220/5220 groups censused, no bit above 18 ever set)
- the macro window is the reference's own `Blizzard_MacroUI`, and `GlobalStrings.lua` — line 1 of the reference's own toc — was missing from our manifest entirely (the app loaded it out of band at startup, so the client was right and every harness that loaded the manifest was not); exactly one `<OnLoad>` runs, the most-derived, because `SetScript` releases the slot before it looks at the new body — which is why the reference re-states the template's width line in tab 1 and not tab 2
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
- the texture filter is the reference's policy — two globals, one process-wide mode, no per-texture override — registered as real CVars every sampler reads: the hardcoded trilinear + aniso 8 at five sampler sites had been read off an apitrace of an install whose `Config.wtf` set `anisotropic "16"`, a configured value taken for a default; and `trilinear` ships ON, because the registrar's `"0"` is not what a fresh install renders at — first-launch hardware detection overwrites sixteen video CVars from its hardware table, both rows a real GPU can reach set trilinear, and the reference's own `gx.log` on this machine names that row — while `anisotropic` is not among the sixteen, so anisotropy stays off
- the MSAA device clamp ran before the config was read, so a saved `gxMultisample = 8` carried from a machine that offers it still killed the render thread on one that stops at 4; the ceiling is now a property of the write itself, covering the config file, `SetCVar`, the dropdown and any writer added later — and an empty device list is no opinion, so headless runs pass through
- one line, one rounding: at a fractional display scale the per-glyph snap's f32 round-trip moved half-pixel ties by an ulp, so caps and ascenders rounded up while x-height letters stayed — the letters left the baseline, the thing the reporter photographed; the line's two anchors now round once and each glyph adds integer device-pixel offsets, so a line cannot be anything but flat at any scale, proven over 15,680 layouts — and `$WOW_DPI`'s size guard, which compared a physical request against the logical size and refused every non-1x run, now reads the unit it is denominated in
- a test that panics while holding `ENV_LOCK` no longer poisons the tests behind it — one missing knob resource had reported as twelve red tests, eleven naming the mutex and none naming the bug; there is nothing for the poison to protect, since the guard's `Drop` restores the environment during the panicking test's own unwind
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
- a tooltip change costs ~1.3 ms — five whole-interface recomputes, each working as designed: the quad re-convert at 474 µs, the Lua tick at 287, the layout resolve at 200, bevy's `Assets<Mesh>` tax at ~270 and the mesh rebuild at 146 — pinned with new hover probes (a jitter dial, a `WOW_PROBE_HOVER_DUTY` duty cycle so run drift cancels, an `AssetChanged` tax meter), and the layout preamble's two whole-roster scans become lists maintained at the writer (52–83 µs → 1–11)
- the splice compared this frame's entries against last frame's by index, so one inserted highlight shifted the 160 entries behind it and read as "everything changed"; it now aligns on `z` — an identity, not a position — and a `Text` entry that writes only quads became spliceable, so a tooltip change re-converts 22 entries instead of 520 (`convert` +439 µs → +107), with `WOW_UI_SPLICE_VERIFY` running the full conversion behind every spliced frame: 0 mismatches over 1,078
- a spell-less macro is a usable action button, not the 0.4 grey — the reference's usable compute has a dedicated leg answering usable for the bound-spell zero, cross-checked at the bytes in the reference's disassembly — and the button's Name line is written unconditionally from `GetActionText` as the reference does, so the macro's name sits under its icon and a `/cast` of a spell the character does not know now greys, which is also 1.12's behaviour (B340)
- a capture bounds its run in wall-clock seconds, because a frame cap is no cap at the ~1 fps a starved `nextDrawable` leaves — 1800 frames is thirty minutes there — and the teardown backstop runs on every run as a liveness guarantee, carrying the exit code the run asked for; the Multisampling row reads Off / 2x / 4x, since "1x" contradicts the no-multisampling byte both of the reference's enumerators normalise to and the bit-depth prefix was identical on every row
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
- the in-game UI ends at the character screen and is rebuilt at the next login, so addons that read the character at file scope stop being frozen on whoever logged in first
- everything the host remembers about a UI is keyed on that VM's identity — keybinds, macro icons and every feed survive the rebuild instead of silently pushing nothing
- a ribbon trail rides the room its model is drawn by: Caverns of Time's ninety energy trails stop burning through 200 yards of rock
- the visibility census gains the sky and ribbon lanes it could not see, and the logout leak probe measures after the release instead of accusing every run
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
- a player build ships beside the dev one — dev chords and probe surfaces go dark, and a gate enforces what a player can still reach, by medium not by module
- the binary carries its own shaders and its shipped UI, finds the WoW folder by itself, and keeps its local state in one benilla-config folder
- the glue screens get a capture shutter, and the engine stops depending on the protocol crate
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
- the world engine splits from the game into a benilla-world crate — 112 files, and the wall is ratcheted to zero names of the game inside the engine
- the engine's public doorway is counted and gated at 32 items, and the render materials, shaders and asset cache move down to benilla-assets
- the game's files leave the engine's asset root — game:// is its own space — and a spike binary boots the engine with no game attached
- weather rides along: snowfall emission is denominated in the reference's frame cap, and a long teleport retires the stranded precipitation field instead of leaving it behind
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
- ctrl-click an item to try it on: the dressing room dresses your double, and shift-click posts the item link to chat
- the pet tab lands in the character window: paper doll stats, family and diet, the right-click menu, and dropping food on the pet feeds it
- water-only spells refuse on dry land: aquatic form gates to swimming, client-side like the real one
- the macro icon picker's white squares get their art back, and option rows explain themselves on hover
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
- every rebindable key runs through one chord-to-command engine, set in the era Key Bindings window, and buttons show their real keys
- a spell aimed at the ground sends its projectile to that spot, arcing to the point instead of a target
- a pet ordered onto something it can't attack refuses cleanly, and its frame shows the happiness face read from the client's own data
- a window's tabs clamp to the drawn edge and highlight exactly where they sit
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
- the pet gets its own unit frame beside yours, and an Attack order no longer latches as the standing command
- five rooms that always read bone dry hold their water now, the Deeprun Tram's flooded stretches among them
- macros fire through the client's own chat door, and the macro window fits its tabs and scrolls its icon chooser
- the options window grows a Nameplates page and its first dropdown, and the name-display switches apply live in the world
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
gromandClaude Opus 5